Vibe coding an engine is way harder than vibe coding something onto an existing engine. It’s something I have worked on for fun in my free time.

I have the ability to make the engine I’m making on my own, but trying out AI for the experience. It really sucks in ways that make it good for what an engine needs. A good engine needs to plan pretty far ahead and plan well at high architecture level. AI is actually awful at that despite it being okay at making plans at implementing said plan.

I didn't try that hard but I did not have much success. I spent some time trying to vibe code a forward clustered renderer in vulkan and I couldn't manage to get anything I was too happy with. Mostly just regurgitation of a few different tutorials. It's possible I'm just too dumb to use AI and it was also 18 months ago, so things have progressed on the LLM front.